> Canonical Huffman's decoding is done in _make_words function. Canonical > Huffman's decoding have one property that, with only codelength > information > we can construct codewords. No. For canonical huffman, you only need the _number_ of words per code length. Vorbis does not use canonical huffman. Segher